TiO2 catalysts by

SURFACE AND INTERFACE ANALYSIS, VOL. 23, 204-212 (1995)

Study of the Reduction Behavior of W/Ti02 Catalysts by XPS using Curve Fitting, Deconvolution and Factor Analysis Joseph N. Fiedor, Andrew Proctor, Marwan Houalla and David M. Hercules* Department of Chemistry, University of Pittsburgh, Pittsburgh, PA 15260, USA

X-ray photoelectron spectroscopy (XPS,ESCA) was used to determine the tungsten oxidation states in a 5 wt.% WO,/l%O, catalyst reduced in hydrogen in the temperature range 400-700 "C.Because, in the region of interest, W 4f overlaps with the Ti 3p energy level, a multiple analysis procedure including deconvolution, non-linear leastsquares curve fitting and factor analysis was used to analyze the data. Deconvolution was better able to resolve the peaks in the W 4f region and thus elucidate the positions of the tungsten components for each reduction temperature. It was shown that there is a maximum of three W 4f doublets. The W 4f,,, peaks of these components were located at 35.5,33.3 and 31.1 eV and they were assigned to W+6, W+4 and W", respectively. Factor analysis verified the presence of three spectral tungsten components. X-ray ditaction verified the presence of W" on reduction at 675OC. Non-linear least-squares curve fitting, using the positions from deconvolution, was able to determine the distribution of tungsten oxidation states as a function of reduction temperature: W+6 steadily decreased with increasing reduction temperature, W" initially appears at 408OC and goes through a maximum around 58OOC; W" was first detected following reduction at 550OC and was the only species present for reduction temperatures higher than 665 OC. From the distribution, it is proposed that, unlike the molybdenum systems, tungsten undergoes a more simplified reduction process in which only one intermediate oxidation state (presumably W+4) is detected.

INTRODUCTION Tungsten-supported catalysts are active for reactions such as hydrogenation, metathesis and isomeri2ati0n.l~~ Activation of these catalysts often requires reduction pretreatment. The reduction step can produce a surface containing a mixture of tungsten oxidation states of varying relevance to the catalytic activity. Therefore, understanding the reduction of supported tungsten catalysts is essential for any attempt to establish surface structure/activity relationships. The reduction of W/TiO, catalysts has been studied . ~Vermaire et ~ 1 using . ~ previously by Bond et ~ 1 and temperature-programmed reduction (TPR). Bond et al. suggested that W + 6 reduces directly to W metal, while Vermaire et al. concluded that W undergoes a two-step reduction (i.e. W + 6+ W + 4+ WO). However, TPR studies, as well as gravimetric and volumetric analyses, are ambiguous because they measure only an average oxidation state. Clearly, there is a need for a more direct method of examining the reduction process and the surface of reduced W/Ti02 catalysts. Recent work has shown that x-ray photoelectron spectroscopy (XPS, ESCA) can be used in conjunction with various data analysis techniques, such as nonlinear least squares curve fitting (NLLSCF) and factor analysis (FA), to determine the distribution of Mo oxidation states in reduced Mo-supported

* Author to whom correspondence should be addressed. CCC 0142-242 1/95/040204-09 0 1995 by John Wiley & Sons, Ltd

The present study will expand on the use of various data analysis methods to define the reduction pathway of a 5 wt.% W03/Ti02 catalyst reduced in hydrogen in the temperature range 400-700 "C. The XPS tungsten energy level investigated in the present study is the W 4f region. The W 4d region proved to be too broad to be unambiguously useful. Analysis of the W 4f region is complicated by interference from the Ti 3p level of the support and other less-intense tungsten core levels. In an attempt to solve this problem, an extensive analysis procedure is undertaken so that the distribution of tungsten oxidation states in reduced 5 wt.% WO,/TiO2 catalysts can be determined with confidence. The use of multiple complementary analysis techniques avoids the ambiguity that often results from using a single technique. First, a deconvolution algorithm' was employed to resolve better the peaks and thus elucidate the number and positions of the W and Ti components for each reduction temperature. Such a procedure has been used previously in the analysis of XPS envelopes,'*' where it was shown to enhance the resolution threefold. Next, a NLLSCF routine used the information provided by the deconvolution procedure concerning the number and positions of the peaks to provide a firmer base for curve fitting the individual W 4f envelope. From NLLSCF results the distribution of W oxidation states in each W 4f envelope was measured. Finally, FA was used to determine the number of tungsten components in the reduction series; the results were compared with those obtained by deconvolution and the NLLSCF routine. Receiued 21 July 1994 Accepted 15 November 1994

REDUCTION BEHAVIOR OF W/TiO, CATALYSTS BY XPS

EXPERIMENTAL AND METHODOLOGY

Catalyst preparation Degussa P-25 TiO, (84/16 anatase/rutile, pore volume = 0.5 cm3 g-' and BET surface area = 50 5 m2 g-') was first mixed with deionized water, dried (110°C) and calcined (400°C) in air for 12 h. It was ground and sieved to 100 mesh, then tungsten (5 wt.% WO,) was deposited by incipient wetness impregnation using an ammonium metatungstate (Cerac) [(NH4)6H2W12040* 4H20] solution. The resulting catalyst was dried at 110°C for 16 h followed by calcination for 16 h in air at 500 "C.

Raman Raman spectra were recorded on a Spex Ramalog spectrometer equipped with holographic gratings. The 5145 A line from a Spectra-Physics Model 165 argon ion laser was adjusted so that 50 mW of power was measured at the sample. The spectral slit width was 4 cm-' and the scan rate was 0.4 cm-' s-'. A Spex datamate computer was used for each data analysis. The peak areas were calculated assuming a linear background. The 5 wt.% W03/Ti02 catalyst and the TiO, support were pressed into 13 mm KBr-backed pellets and scanned under ambient conditions. The samples were rotated to minimize local heating. Raman was used to determine if the 5 wt.% W03/Ti02 catalyst contained any bulk tungsten compounds (e.g. WO,).

X-ray difiaction A Diano Model 700 diffractometer using Ni-filtered Cu Kcr radiation (A = 1.540 456 A) equipped with a graphite monochromator was used to obtain x-ray diffraction (XRD) data. The copper x-ray tube was operated at 50 kV and 25 mA, and the spectrum was scanned at a rate of 0.4" min-' (in 28). The pressed catalysts used in the reduction treatment for XPS analysis (see XPS section for reduction procedure) were mounted on a glass slide for XRD analysis. The peak areas of anatase (101) and rutile (110) with reflections at 48.1" and 41.2" (in 28), respectively, were measured to determine the anatase/ rutile composition of the reduced catalysts. An equation relating the intensity ratio for these reflections to per cent anatase was obtained from Ref. 11.

205

data analysis. Binding energy values for the 5 wt.% W03/Ti0, catalysts were referenced to the Ti 2p3,, line of TiO, (458.7 eV',). For reduction experiments, the catalysts were pressed (2000 psi) into pellets and mounted on a sealable probe.', The sample was heated to the desired temperature (400-700 "C)by ramping the temperature at a rate of 2" min-' in flowing H, at a flow rate of 100 cm3 min-'. The catalysts were reduced for 12 h at the desired reduction temperature and then cooled to room temperature under H, flow. The sealable probe was pressurized to 20 psi, sealed and transferred to the spectrometer.

Deconvolution Deconvolution was accomplished by using the point simultaneous over-relaxation Jansson a l g ~ r i t h m . ~ . ' ~ Deconvolution is the removal of broadening effects from a spectrum. An XPS spectrum is broadened by convolution of the intrinsic signal with several broadening functions that are either Gaussian or Lorentzian in nature. Examples of these broadening functions include the natural linewidth (due to the lifetime of the corehole state), the exciting x-ray lineshape, the detection system and any broadening due to charging. In order to achieve a successful deconvolution, each spectrum must be pretreated. Pretreatment includes background removal and spectral smoothing. The backgrounds were removed using a Shirley-type integral14 and spectral smoothing was carried out using a cubic All data analysis programs (GOOGLY Software) were written in house by A.P. The Jansson algorithm implements an iterative type of procedure that can be controlled interactively by a visual evaluation or by monitoring the residual variance between the original data and the reconstructed data (i.e. the convolution of the broadening function and the current deconvoluted spectrum). The most important variable in any deconvolution is the broadening function. In the present case the broadening function was chosen to be a symmetric Voigt function" with 20% Lorentzian character. The width chosen happened to be slightly narrower than those used in curve fitting (see below) the spectrum of interest. In our case the width was selected to be 1.9 eV. This allows for maximal deconvolution while not losing components because of 'over deconvoluting'. The deconvolution procedure is also able to provide information about the peak positions of the Ti and W components as well as the number of spectral components for each reduction temperature. Thus, it can give an indication when each new tungsten component begins to emerge in the reduction series.

X-ray photoelectron spectroscopy The XPS spectra were obtained with a modified AEI ES2OOA photoelectron spectrometer equipped with an A1 anode (hv = 1486.6 ev) at a power of 240 W (12 kV, 20 mA). The instrument was operated in the fixed retardation ratio (FRR) mode and the analysis pressure was < 5 x lo-' Torr. The instrument was interfaced to an IBM PC compatible for data collection and subsequent

Non-linear least-squares curve fitting The methodology used to fit the individual Ti 3 p W 4f envelopes was slightly different than that adopted in previous work',' because of the Ti 3p interference. Information about the Ti 3p region for the W/Ti02 catalysts, such as peak width and peak position, was

J. N. FIEDOR ET AL.

206

obtained by curve fitting the Ti 3p region of TiO, alone. Once the Ti 3p envelope was fitted, the peak position and peak width were kept relatively constant (i.e. 50.2 eV) when fitting the envelopes of the reduced catalysts. This can be done because there is no evidence from XPS for any significant change of the Ti 2p envelope as a function of reduction temperature. Therefore, it is safe to assume that the Ti 3p envelope will not change either. Curve fitting of the W 4f envelopes was accomplished by using the Levenberg-Marquardt damping method.” All peaks were fitted using a Voigt function’* with 20% Lorentzian character. The background was assumed to be integral and it was applied individually to each peak. The W 4f5,,/W 4f7/2 area ratio for the spin-orbit doublet was fixed at 0.78.20The separation between the W 4f7/2 and W 4f5,, was set at 2.1 eV and the width ratio was assumed to be unity. Figure 1 shows the W 4f region for WO,. It can be seen that a small high-binding-energy peak (indicated by an arrow) appears at - 6 eV higher than the main W 4f,/, peak. A similar high-binding-energy peak was reported earlier” for W metal and tentatively assigned to the W 5p3/2 level. To verify this hypothesis a highbinding-energy window was selected to include both W 5p components. The small broad peak centered around 50 eV corresponds to the W 5p,/, energy level. The W 5p3/2 peak should be located at -12 eV from the W 5p,,, energy and it should exhibit a similar full width at half-maximum (FWHM = 4.5 eV). This would put it more or less into the spectral background of W 4f. As the additional high-binding-energy peak in the W 4f region is clearly narrower and its position does not coincide with the theoretical value, it cannot be attributed to the W 5p3/, peak. In our fitting procedure the small high-energy peak was designated as a ‘satellite’ and fitted as a single peak. Figure 2 shows the fit for W 0 3 . The position of the ‘satellite’ peak relaxed to 41.8 eV. An effort was made to fit the W peak but its area completely disappeared when it was allowed to float. Therefore, it was assumed that the W 5p,,, area is contained within the individual integral background. The area ratio and the binding energy difference between the ‘satellite’ peak and the main W 4f peak were kept constant when fitting the

L

\--

L-

~

p

_

60

_

L

-

50

Binding Energy I eV Figure 1. Spectrum of WO,

.

A - I I 40 30 ~

A

A-

r”\ REX%=6.0

i

“satellite”

Binding Energy I eV Figure 2. Curve-fitted spectra of WO,

envelopes of the reduced catalysts. As the high-bindingenergy peak and the W 5p3/, peak contribute only marginally to the Ti 3 p W 4f envelope, any errors associated with it will not affect the overall results. The next step was to fit the envelopes of the reduced catalysts. The tungsten parameters used for these envelopes were the same as those established for WO, . As mentioned previously, the Ti 3p peak width and peak position were kept relatively constant throughout the reduction series. The initial peak positions of the tungsten components were those obtained by deconvolution. However, when performing NLLSCF these tungsten positions were allowed to float until they relaxed into their local minimum.

Spectral subtraction A difference procedure was implemented to remove the Ti 3p contribution from the Ti 3 p W 4f envelope so that subsequent FA could be performed. When performing a difference procedure one must ensure that the two spectra are correctly aligned and n o r m a l i ~ e d In .~~ our case the two spectra used in the analysis are the Ti 3p envelope of the TiO, support and the Ti 3 p W 4f envelopes of the reduction series (see Fig. 6). To begin the alignment and/or calibration procedure, each spectrum must undergo interpolation. A detailed description of the interpolation procedure can be found in Ref. 23. Once each spectrum is interpolated, the two spectra are calibrated by referencing each spectrum to the Ti 2p,,, line of TiO, (458.7 eVI2). On that basis, the Ti 3p peaks for both the support and 5 wt.% W03/Ti0, catalysts are positioned at 37.3 eV. Next, one must carry out a normalization process. In the present case normalization consisted of determining the Ti 3p area to be removed from the Ti 3 p W 4f envelope. The normalization constant ( N C ) for each reduction temperature is simply


where (Ti 2p),,, and (Ti 2p),,, are the Ti 2p areas for the catalysts and the TiO, support alone. The Ti 3p peak of the TiO, support was then multiplied by the normalization constant and the subsequent subtraction was carried out.

207

.= .

. I

Factor analysis The requirements for FA and a detailed description of the technique have been given earlier.6924In this study, F A comprised only principal component analysis (PCA), which involves the decomposition of a data matrix consisting of c spectra into c abstract (non-spectroscopic) components by single value decomposition (SVD).I9 Through PCA it is possible to determine which abstract components account for signal (1 . .. n) and which account for noise (n + 1 .. . c). This study will focus on the weight of evidence of several statistical methods, such as the F indicator function IND24 and the reduced eigenvalue (REV) ratio,24 as well as other non-statistical methods, to determine n. A more detailed explanation for the determination of n by these various techniques can be found in Refs 24 and 27. None of the criteria chosen for this work require prior knowledge of the noise content in the original data matrix.24

RESULTS Raman Raman spectroscopy was used to determine if the 5 wt.% W03/Ti02 catalyst contained any bulk tungsten compounds (e.g. W03). The Raman spectrum (not shown) of the 5 wt.% WO,/TiO, catalyst showed no peaks characteristic of W 0 3 . A broad band at -960 cm-' was observed and is attributed to the W=O stretching band of the tungsten interaction species.,* These Raman results indicate that the tungsten phase in the 5 wt.% WO,/TiO, catalyst is exclusively present as a surface tungsten species. X-ray difiaction X-ray diffraction analysis of the reduced 5 wt.% W03/Ti02 catalysts was performed to monitor any changes in the anatase/rutile composition and the formation of any new tungsten phases. Figure 3 plots the anatase/rutile ratio as a function of reduction temperature for the 5 wt.% W0,/Ti02 catalyst. Samples reduced at temperatures 600 "C exhibit a constant anatase/rutile composition that is approximately equal to that of the oxidic catalyst. A sharp decline in the anatase/rutile ratio occurs for the samples reduced between 600 and 712°C. This type of behavior was also observed for the Mo/TiO, system.' The same study also found that the TiO, support alone undergoes the change from anatase to rutile at a much lower temperature, indicating that Mo inhibits the transformation of anatase to rutile upon reduction (7). Our data shown in Fig. 3 indicate that the W/TiO, system behaves in a similar manner. These results agree with those postulated by Ramis et al.29

-=

Reduction Temperature (/OC) Figure 3. Anatase/rutile ratio of the TiO, support as a function of reduction temperature.

An in situ XRD experiment was carried out to verify the presence of W metal for catalysts reduced at -700°C. For this experiment a lo00 mg catalyst was packed against a quartz wool plug in a quartz reactor tube. The heating, reduction and cooling procedures were the same as those used in the XPS reduction experiments. After the reactor was cooled it was sealed and the samples were transferred to a glove box containing ultrahigh purity Ar. In the glove box the samples were packed into a 2 x 2 x 0.2 cm hollowedout plastic slide and covered with Kapton tape so that the XRD measurements could be taken with minimal exposure of the reduced catalysts to air. Figure 4 shows the XRD reflections in the region of 30-50" in 26 for the TiO, support (a), the support reduced at 675°C (b), W metal (c) and a 5 wt.% WO,/TiO, catalyst reduced at 675 "C (d). The oxidic TiO, support [Fig. qa)] shows XRD reflections characteristic of anatase and rutile. The reflections characteristic of anatase are located in 28 at 36.9", 37.8", 38.5" and 48.07", while those characteristic of rutile are found at 36.05", 39.2", 41.2" and 44.05" in 26. It is evident from the relative intensity of these reflections that anatase is the dominant form present in the oxidic T i 0 2 support. Upon reduction to 675 "C the reflections characteristic of anatase disappear and only those of rutile remain [Fig. 4(b)]. The XRD pattern of W metal exhibits an intense reflection at 40.2" in 28 [Fig. 4(c)]. Figure 4(d) displays the XRD spectrum of the 5 wt.% W03/Ti02 catalyst reduced at 677°C. In addition to the lines characteristic of rutile, the XRD pattern of the reduced catalyst shows a broad shoulder at 40.2" which can be attributed to W metal.

-

-

X-ray photoelectron spectroscopy X-ray photoelectron spectroscopy was used to provide information about the tungsten dispersion and distribution of tungsten oxidation states of the 5 wt.% W03/Ti02 catalyst as a function of reduction temperature. Because of the overlap between W 4f and Ti 3p, the W 4d region was used for monitoring tungsten dispersion. Figure 5 shows the variation in the W 4d/Ti 2p XPS intensity ratio as a function of reduction temperature. The oxidic catalysts and the catalysts reduced between 400 "C and 500 "C exhibit similar intensity

J. N. FIEDOR ET AL.

208

Reduction Temp ( P C )

oxidic 408 503 529 549 582 604 629 650 662 712 I

32

36

44

40

40

2 theta

Figure 4. X-ray diffraction reflections for: (a) TiO,; (b) TiO, reduced at 675°C; (c) W metal powder; (d) a 5 wt.% WO,/riO, catalyst reduced at 677 "C.

1

0.4 I

a

0.3

r\l

O

~

400

9

'

"

"

.

' " 500

"

'

" ' 600

'

I

I

I

36 32 Binding Energy I eV

Figure 6. The Ti 3p-W 4f spectra for re1 iced 5 wt. catalysts.

WOfliO,

(indicated by arrows) emerge. However, because the Ti 3p energy level is a major contributor to these XPS envelopes it is difficult to determine the exact nature of the tungsten components as a function of reduction temperature. Therefore, there is a need to improve the resolution in a manner that will help to define the positions of all the tungsten components. Thus, a deconvolution procedure was carried out on each individual Ti 3 p W 4f envelope in an attempt to resolve the components. Figure 7 displays the series of deconvoluted Ti 3 p W 4f envelopes. A decrease in peak width by about a factor of 3 can be seen. The peak (A) at 37.5 eV attrib-

700

Reduction Temperature ( P C ) Reduction Temp. ( P C )

Figure 5. Variation of ESCA W 4d/Ti 3p intensity ratio as a function of reduction temperature.

oxidic 408

ratios. A considerable decrease in the intensity ratio is observed for reduction temperatures between 500 "C and 600 "C. This can be attributed to particle growth of the tungsten surface species and/or a change in the morphology of the surface tungstates upon Note the increase in the intensity ratio from 650°C to 700 "C. This behavior has been observed previously for the Mo/TiO, system and it can be attributed to the collapse of the TiO, surface area, which occurs on transformation from anatase to rutile.

503

529 549 582 604 629

DATA ANALYSIS

650 662

Deconvolution Figure 6 shows the series of W 4f-Ti 3p envelopes obtained from a set of reduced 5 wt.% WOJTiO, catalysts. It is clear that as the reduction temperature increases from 400 to 700°C new tungsten components

I

40

I

I

I

712


Figure 7. Deconvoluted Ti 3p-W 4f envelopes for reduced 5 catalysts.

wt.% WO,/TiO,


uted to Ti 3p is constant throughout the series. As the reduction temperature is increased it is evident that new tungsten components appear. These new components are indicated by arrows. The four positions appearing in the reduction series are 37.5 eV (A), 35.5 eV (B), 33.3 eV (C) and 31.1 eV (D), respectively. The peaks at 35.5 eV and 31.1 eV are assigned to the W 4f,/2 positions for W+6 and W", respectively, based on model compounds. W 4f7/2 binding energy values ranging from 34.5 eV to 32.7 eV have been assigned to W+4.30-35On this basis the peak at 33.3 eV can be assigned tentatively to the W 4f7/2 position of W+4. Close examination of the individual deconvoluted spectra (Fig. 8) reveals that the peak assigned to W+4 appears at 503 "C and that the peak assigned to Wo first appears at 549°C. From Fig. 7 one can see that as the reduction temperature is increased the peak assigned to W 4f7/2 of W + 6 disappears (-650°C) and the peak assigned to W 4f7/2 of W" increases. The behavior of the W + 4 species is not quite as straightforward. This is partly due to the fact that the doublet separation (2.1 ev) of a particular tungsten oxidation state coincides with the separation between the W 4f7,, peaks of W f 6 and W + 4 (i.e. 2.2 eV) and that of W'4 and W" (i.e. 2.2 eV). Non-linear least-squares curve fitting

209

cedure on the original experimental data the positions were allowed to relax slightly into their local minima. Typical examples of curve-fitted Ti 3 p W 4f envelopes of the catalyst are shown in Fig. 9. Figure 9 shows the Ti 3 p W 4f envelopes (*) for the 5 wt.% W03/Ti02 catalyst reduced at 529 "C and at 582 "C. for The residuals indicate that the overall fits (-) both envelopes agreed well with the experimental data. Statistically, the goodness of fit was measured by the residual excursion (REX%) value.6 The lower the REX% values the better the fit. The REX% values are noted on the figure. Figure 9(a) has a REX% of 4.1 while the value for Fig. 9(b) is 3.8. Both values indicate relatively good fits. Because the fitting procedure used in this study is complicated, it is important to show that the fits are legitimate. To verify the legitimacy of the fits certain experimental criteria were used. First, results from the fits show that the Ti 3p fitted area was found to be within f 7 % of the calculated Ti 3p area. The calculated Ti 3p peak area was determined by multiplying the measured Ti 3p/Ti 2p ratio for TiO, alone by the measured area of the Ti 2p region for the catalysts. Another way to check the consistency of the fits was to compare the variation of W 4f/Ti 2p intensity ratios with those of W 4d/Ti 2p (see Fig. 5) as a function of reduction temperature. The W 4f area was found by summing the fitted area of the W 4f doublets for each

A NLLSCF routine was used to determine the distribution of tungsten species as a function of reduction temperature. Deconvolution has provided a logical starting point for the NLLSCF routine. It was able to give positional information for all the components and to provide an indication of the reduction temperatures at which new components appear. Thus, the initial guesses used in the NLLSCF routine are more accurate, which inherently makes the curve fitting procedure a more efficient process. In performing the NLLSCF pro-

V

REX%

A

-

4.1

W4f-Ti3p

I

40

I

I

36

32

Binding Energy I eV

REX% = 3.8

44

I


40

Figure 8. Deconvoluted spectra for 5 wt.% WO,piO, reduced at 503 "C and at 549 "C.

40

I

I

36

32

Binding Energy I eV

catalysts

Figure 9. Curve fits for 5 wt.% WO,/TiO,

529°C and (b) 582°C.

catalysts reduced at (a)

J. N. FIEDOR ET AL.

210

Reduction Temp. (PC) Figure 10. Distribution of tungsten oxidation states in 5 wt.% WO,/TiO, catalysts as a function of reduction temperature.

oxidation state at each reduction temperature. The W 4f energy level gave a similar trend (not shown) to that observed in Fig. 5. Figure 10 shows the distribution of tungsten oxidation states as a function of reduction temperature: W + 6 decreases steadily until it disappears at 650 "C; W + 4 increases steadily with increasing reduction temperature, reaches a maximum at -529"C, stays constant until 629 "C and then decreases for higher reduction temperatures; and Wo first appears at 549 "C, which is in agreement with the deconvolution results, and rapidly increases until 712°C where it is the only species detected.

that the analysis will provide information about only the spectral tungsten components. The envelopes of the reduced catalysts were used as the original data matrix for PCA. The W 4f envelope of the oxidic catalyst was not included in the analysis because this sample exhibited significant charging and therefore will probably experience charge broadening compared to the others. This may cause PCA to overestimate the number of components and/or factors in the original data matrix. As the reduction temperature is increased one can readily observe that the envelope broadens owing to the appearance of other tungsten oxidation states. Principal component analysis was performed on this series of envelopes to determine the number of tungsten oxidation states or components in the original overall data matrix (Fig. 11). Data pretreatment included removal of the background from each spectrum by using a Shirleytype integral,I4 subtracting the minimum from each spectrum and scaling the area of each to a constant value. Initial PCA results are given in Table 1. All criteria indicate the presence of three factors. The IND function minimizes at the third factor, the REV ratio rises from 1.47 to 7.40 on moving from factor 4 to factor 3, and the Q% value falls below 10 from factor 4 to factor 3. Therefore, all the statistical criteria quite convincingly point to the fact that the original data matrix is comprised of three factors. This result is consistent with those obtained by deconvolution and NLLSCF. It should also be pointed out that if the Ti 3p peak is included, then the rank of the data matrix merely increases by one. This is to be expected. DISCUSSION

Factor analysis Figure 11 shows the W 4f envelopes from the series of reduced 5 wt.% W0,/Ti02 catalysts following subtraction of the Ti 3p envelope. The Ti 3p was removed so Reduction Temp. (/OC)

408 503

529 549

As mentioned previously, few studies have focused on the reduction of W/Ti02 catalysts. Temperatureprogrammed reduction experiments conducted by Bond et al. showed that a monolayer W/Ti02 catalyst prepared by incipient wetness impregnation reduces directly and completely to Wo between the temperatures of 730 and 770"C.4 They also pointed out that above monolayer coverage (4.8 wt.% WO,) the presence of bulk WO, suppresses the reduction to Wo. These results were at variance with those reported by Vermaire and van Berge5 which suggest that a monolayer WO,/TiO, catalyst prepared by equilibrium adsorption undergoes a two-step reduction (W+6+ W t 4 + WO).

582 604 629 650 662 712 I

I

I

35 30 Binding Energy / eV

40

Figure 11. Original data matrix for reduced 5 wt.% WO,/TiO, catalyst from which the Ti 3p peak has been removed.

Table 1. Principal component analysis results for the data matrix shown in Fig. 11 Factor

1

2 3 4 5 6 7 8 9 10

% Eigenvalue

REV ratio

IND

0%

79.467 19.683 0.517 0.063 0.037 0.022 0.013 0.008 0.006 0.004

3.62 33.73 7.1 0 1.47 1.35 1.39 1.19 0.92 0.73 1.o

4.1 1 E-04 1.00E-04 6.69E-05 7.53E-05 9.14E-05 1.21E-04 1.89E-04 3.86E-04 1.39E-04 0.00

0.24 0.48 1.12 19.78 24.76 29.45 37.57 45.96 54.84 40.45


-

The TPR peak associated with the formation of W + 4 occurs at 744 "C; that corresponding to W" formation occurs around 975°C. They also stated that the tungsten surface species did not reduce completely to tungsten metal. Their TPR results indicated that 32% of the tungsten remained as W+4 at 1227°C. This was attributed to stabilization of the W-0 bond of WO, by the rutile structure of TiO, . In a sense, our results agree with some part of both works. For instance, XPS results suggest that the tungsten surface phase of a 5 wt.% W03/Ti02 catalyst reduces completely to Wo at 712°C. The in situ XRD results verify the presence of Wo (i.e. reflection at 40.2') following reduction at 677°C. The in situ XRD results also showed that there were no reflections characteristic of WO, at this reduction temperature. This finding is more consistent with that of Bond et al. However, it is clear from the present work that there are three tungsten components, assigned to W f 6 , W f 4 and W", present in the reduction temperature range of 400700 "C. This is consistent with the results found by Vermaire and van Berge.5 It is interesting to compare the reduction results of the W/TiO, system with those reported for supported Mo catalysts. Previous XPS studies of Mo/TiOZ7 and Mo/Al,O,* indicated that Mo forms a wide range of stable oxidation states (i.e. M o + ~M , o + ~M , o + ~M , o+~, Mo" and Moo) upon reduction with H, . Fiedor et aL6 also showed that three Mo oxidation states (i.e. M o + ~ , M o + ~and M o + ~ )were present on a 5 wt.% MoO,/TiO, catalyst after reduction at relatively low temperatures (< 300 "C). The results of the present study for a monolayer-like W/TiO, system indicate that, unlike the Mo systems, tungsten undergoes a more simplified reduction process in which only one intermediate oxidation state was detected. It should be noted, however, that the presence of an intermediate tungsten oxidation state, such as W +,,in relatively low concentrations or having a binding

211

energy value very close to that of W + 4 or tungsten metal cannot be completely ruled out. Non-linear leastsquares curve fitting could conceivably miss the contribution from a minor component if its presence does not better the overall fit. Factor analysis can also miss minor constituents of the overall data matrix. This should not significantly affect the main conclusion of this study; namely, that unlike the Mo system only one important intermediate tungsten oxidation state can be detected on the reduction of monolayer-like W/Ti02 catalysts.

CONCLUSION

A detailed analysis procedure, which included deconvolution, NLLSCF and FA, of XPS data from a series of reduced 5 wt.% W03/Ti02 catalysts has indicated that three tungsten components are needed to describe the data. These tungsten components have been assigned to W+6,W + 4and W". X-ray diffraction verified that tungsten reduces to W" at 675 "C. It was shown that W + 6 decreases with increasing reduction temperature; W + 4 first appears around 400 "C, goes through a maximum around 580 "C and then decreases as the reduction temperature is increased further, W" first appears around 550 "C then maximizes at -675 "C at 100%. It is concluded that, unlike the Mo/TiO, system, the tungsten in a monolayer-like W/TiO, catalyst undergoes a more simplified reduction process in which only one intermediate oxidation state was detected. N

Acknowledgement The authors acknowledge financial support for this work from the US Department of Energy, Grant DE-FG02-87ER13781.

REFERENCES 1. R. Thomas, E. M . van Oers, V. H. J. De Beer, J. Medema and J. A. Moulijn, J. Catal. 76, 241 (1982). 2. W. Grunert. R. Feldhaus, K. Anders. E. S. Shpiro and Kh. M. Minachev, J. Catal. 120,444 (1989). 3. T. Yamaguchi, Y. Tanaka and Y. Tanabe, J . Catal. 65, 442 (1980). 4. G. C. Bond, S. Flamerz and L. Van Wijk, Catal. Today 1, 229 (1987). 5. D. C. Vermaire and P. C. van Berge, J. Catal. 116, 309 (1989). 6. J. N. Fiedor, A. Proctor, M . Houalla and D. M . Hercules, Surf. Interface Anal. 20. 1 (1 993). 7. R. B. Quincy, M. Houalla, A. Proctor and D. M . Hercules, J. Phys. Chem. 94,1520 (1 990). 8. M . Yamada. J. Yasumaru, M. Houalla and D. M . Hercules, J. Phys. Chem. 95,7037 (1991 ). 9. P. A. Jannson, Deconvolution. Academic Press, Orlando (1984). 10. D. Sprenger and 0. Anderson, fresenius J . Anal. Chem. 341, 116 (1991). 11. R. A. Spurr and H. Myers, Anal. Chem. 29, 760 (1957). 12. R. B. Quincy, M . Houalla, A. Proctor and D. M . Hercules, J. Phys. Chem. 93,5882 (1989). 13. K. T. Ng and D. M . Hercules, J. Phys. Chem. 80, 2094 (1976). 14. D. A. Shirley, Phys. Rev. B 5, 4709 (1972). 15. C. R. Reinsch, Numer. Math. 10, 177 (1967).

16. C. A. deBoor, A Practical Guide to Splines. Springer, New York (1978). 17. A. Proctor, M. J. Fay, D. H. Hoffmann and D. M. Hercules, Appl. Spectrosc. 44, 1052 (1990). 18. J. J. Humlicek, Quant. Spectrosc. Radiat. Trans. 21, 309 (1979). 19. W. H. Press, B. P. Flannery, S. A. Teukolsky and W. T. Vetterburg, Numerical Recipes. Cambridge University Press, Cambridge (1986). 20. J. H. Scofield, J. Electron Spectrosc. Relat. Phenom. 8, 129 (1976). 21. T. E. Madey, J. T. Yates and N. E. Erikson, Surf. Sci. 43, 526 (1974). 22. C. D. Wagner, W. M . Riggs, L. E. Davis and J. F. Moulder, Handbook of X-ray Photoelectron Spectroscopy, ed. by G . E. Mullenberg. Perkin-Elmer Corp., Eden Prairie, M N (1979). 23. J. N. Fiedor, A. Proctor, M. Houalla, P. M. A. Sherwood, F. M. Mulcahy and D. M. Hercules, J. Phys. Chem. 96, 10967 (1992). 24. E. R. Malinowski, factor Analysis in Chemistry, 2nd Edn. Wiley, New York (1 991 ). 25. E. R. Malinowski, J. Chemom. 3, 49 (1988). 26. E. R. Malinowski, J. Chemom. 4, 102 (1990). 27. P. J. Gemperline and J. C. Hamilton, in Computer-enhanced Analytical Spectroscopy, Vol. 2, ed. by H. L. C. Meuzelaar.

212


Plenum Press, New York (1990). 28. R. B. Quincy, M. Houalla and D. M. Hercules, unpublished results (1992). 29. G. Rarnis, G. Busca, C. Cristiani, L. Lietti, P. Forzatti and F. Bregani, Langmuir 8,1744 (1992). 30. W. Grunert, E. S. Shpiro. R. Feldhaus, K. Anders. G . V. Antoshin and Kh. M. Minachev, J . Catal. 107, 522 (1987). 31. P. Biloen and G. T. Pott, J. Catal. 30,169 (1973).

32. J. Haber, J. Stoch and L. Ungier, J. Solidstate Chem. 19, 113 (1976). 33. R. J. Colton and J. W. Rabalais, lnorg. Chem. 15, 237 (1976). 34. L. Salvati, L. E. Makovsky, J. W. Stencel, J. R. Brown Jr and D. M. Hercules,J. Phys. Chem. 85,3700 (1981). 35. I. E. Wachs, C. C. Cherisch and J. H. Hardenburgh, Appl. Catal. 13,335 (1985).